diff --git a/.github/workflows/ci.yml b/.github/workflows/ci.yml
index 99cf83c2c6..c9b7064643 100644
--- a/.github/workflows/ci.yml
+++ b/.github/workflows/ci.yml
@@ -23,7 +23,7 @@ jobs:
     strategy:
       matrix:
         os: [ubuntu-latest]
-        scala: [3.4.2]
+        scala: [3.3.3]
         java: [temurin@18]
     runs-on: ${{ matrix.os }}
     steps:
@@ -56,7 +56,7 @@ jobs:
     strategy:
       matrix:
         os: [ubuntu-latest]
-        scala: [3.4.2]
+        scala: [3.3.3]
         java: [temurin@18]
     runs-on: ${{ matrix.os }}
     steps:
diff --git a/common/shared/src/main/scala/org/specs2/data/LruCache.scala b/common/shared/src/main/scala/org/specs2/data/LruCache.scala
new file mode 100644
index 0000000000..0de39b53f0
--- /dev/null
+++ b/common/shared/src/main/scala/org/specs2/data/LruCache.scala
@@ -0,0 +1,40 @@
+package org.specs2.data
+
+import org.specs2.fp.*
+import org.specs2.control.*
+import org.specs2.time.*
+
+/** LRU (least recently used) cache for processing items Values can be registered and the cached cleaned so that it
+  * doesn't go above a given size. The oldest elements are removed first.
+  */
+class LruCache[A](maxSize: Int, systemTime: SystemTime = JavaSystemTime):
+  private var values: Map[A, Long] = Map.empty
+
+  /** Checks if a value has already been processed; if not immediately adds it to the cache. If it has been processed,
+    * refresh its timestamp.
+    * @return
+    *   the processed status
+    */
+  def register(value: A): Operation[ProcessedStatus] =
+    Operation.delayed {
+      this.synchronized:
+        val alreadyProcessed = values.contains(value)
+        // refresh the timestamp even if the params were already registered
+        values += value -> systemTime.nanoTime
+        val status = if alreadyProcessed then ProcessedStatus.AlreadyProcessed else ProcessedStatus.ToProcess
+        while values.size > maxSize do values -= values.minBy(_._2)._1
+        status
+    }
+
+  /** Return the number of elements in the cache */
+  def size: Int =
+    values.size
+
+  /** Return the timestamp for the oldest element */
+  def oldestTimestamp: Long =
+    values.minBy(_._2)._2
+
+/** This enum describes the status of an item in the LruCache */
+enum ProcessedStatus:
+  case AlreadyProcessed
+  case ToProcess
diff --git a/common/shared/src/main/scala/org/specs2/io/FileSystem.scala b/common/shared/src/main/scala/org/specs2/io/FileSystem.scala
index fef8a24e46..7aa1883a8a 100644
--- a/common/shared/src/main/scala/org/specs2/io/FileSystem.scala
+++ b/common/shared/src/main/scala/org/specs2/io/FileSystem.scala
@@ -3,6 +3,8 @@ package io
 
 import control.*
 import fp.syntax.*
+import data.LruCache
+import data.ProcessedStatus.*
 import java.io.*
 import java.util.regex.Pattern.*
 import java.util.regex.Matcher.*
@@ -58,6 +60,28 @@ case class FileSystem(logger: Logger) extends FilePathReader:
   def mkdirs(path: FilePath): Operation[Unit] =
     mkdirs(path.dir)
 
+  /** Unjaring the same thing over and over is inefficient. LRU cache to keep track of what was already done. */
+  private val UnjarLRUCache = new LruCache[(URL, DirectoryPath, String)](maxSize = 1000)
+
+  /** Unjar the jar (or zip file) specified by "path" to the "dest" directory. Filters files which shouldn't be
+    * extracted with a regular expression. This is only done once per argument list (unless eventually evicted from LRU
+    * cache).
+    * @param jarUrl
+    *   path of the jar file
+    * @param dest
+    *   destination directory path
+    * @param regexFilter
+    *   regular expression filtering files which shouldn't be extracted; the expression must capture the path of an
+    *   entry as group 1 which will then be used relative to dirPath as target path for that entry
+    * @see
+    *   [[unjar]]
+    */
+  def unjarOnce(jarUrl: URL, dest: DirectoryPath, regexFilter: String): Operation[Unit] =
+    for
+      status <- UnjarLRUCache.register((jarUrl, dest, regexFilter))
+      _ <- unjar(jarUrl, dest, regexFilter).when(status == ToProcess)
+    yield ()
+
   /** Unjar the jar (or zip file) specified by "path" to the "dest" directory. Filters files which shouldn't be
     * extracted with a regular expression.
     * @param jarUrl
@@ -67,6 +91,9 @@ case class FileSystem(logger: Logger) extends FilePathReader:
     * @param regexFilter
     *   regular expression filtering files which shouldn't be extracted; the expression must capture the path of an
     *   entry as group 1 which will then be used relative to dirPath as target path for that entry
+    *
+    * @see
+    *   [[unjarOnce]]
     */
   def unjar(jarUrl: URL, dest: DirectoryPath, regexFilter: String): Operation[Unit] =
     val regex = compile(regexFilter)
diff --git a/common/shared/src/main/scala/org/specs2/time/SystemTime.scala b/common/shared/src/main/scala/org/specs2/time/SystemTime.scala
new file mode 100644
index 0000000000..4a94213107
--- /dev/null
+++ b/common/shared/src/main/scala/org/specs2/time/SystemTime.scala
@@ -0,0 +1,9 @@
+package org.specs2.time
+
+/** This trait provides the current time */
+trait SystemTime:
+  def nanoTime: Long
+
+object JavaSystemTime extends SystemTime:
+  override def nanoTime: Long =
+    System.nanoTime()
diff --git a/html/src/main/scala/org/specs2/reporter/HtmlPrinter.scala b/html/src/main/scala/org/specs2/reporter/HtmlPrinter.scala
index a079d60296..95024ce625 100644
--- a/html/src/main/scala/org/specs2/reporter/HtmlPrinter.scala
+++ b/html/src/main/scala/org/specs2/reporter/HtmlPrinter.scala
@@ -187,7 +187,7 @@ case class HtmlPrinter(env: Env, searchPage: SearchPage, logger: Logger = Consol
       case Some(url) =>
         val fs = env.fileSystem
         if url.getProtocol.equalsIgnoreCase("jar") then
-          fs.unjar(jarOf(url), outputDir, s"^${quote(base.path)}(/${quote(src.path)}/.*)$$")
+          fs.unjarOnce(jarOf(url), outputDir, s"^${quote(base.path)}(/${quote(src.path)}/.*)$$")
         else fs.copyDir(DirectoryPath.unsafe(url.toURI), outputDir / src)
       case _ =>
         val message = s"no resource found for path ${(base / src).path}"
diff --git a/tests/shared/src/test/scala/org/specs2/data/LruCacheSpec.scala b/tests/shared/src/test/scala/org/specs2/data/LruCacheSpec.scala
new file mode 100644
index 0000000000..c94cdbeea1
--- /dev/null
+++ b/tests/shared/src/test/scala/org/specs2/data/LruCacheSpec.scala
@@ -0,0 +1,56 @@
+package org.specs2
+package data
+
+import org.scalacheck.*
+import org.scalacheck.Arbitrary.*
+import org.specs2.time.*
+import org.specs2.fp.syntax.*
+import ProcessedStatus.*
+
+class LruCacheSpec extends Specification with ScalaCheck:
+  def is = s2"""
+
+  A LRU cache can be used to store elements and evict them when they have been unused for a long time
+    A status is returned to know if an element has already been seen before $e1
+    The cache can not contain more than a fixed number of elements $e2
+    The oldest elements are always evicted first $e3
+
+  """
+
+  def e1 =
+    val cache = LruCache[Int](maxSize = 3, systemTime = MockSystemTime())
+    val operations = cache.register(1) >> cache.register(2) >> cache.register(1)
+    val status = operations.unsafeRun
+    status === AlreadyProcessed
+
+  def e2 = prop { (n: SmallInt) =>
+    val cache = LruCache[Int](maxSize = 3, systemTime = MockSystemTime())
+    val operations = (1 to n.value).toList.traverse(i => cache.register(i))
+    operations.void.unsafeRun
+    cache.size must be_<=(3)
+  }.set(minTestsOk = 10)
+
+  def e3 = prop { (n: SmallInt) =>
+    val mockSystemTime = MockSystemTime()
+    val cache = LruCache[Int](maxSize = 3, systemTime = mockSystemTime)
+    val operations = (1 to n.value).toList.traverse(i => cache.register(i))
+    operations.void.unsafeRun
+    cache.oldestTimestamp must be_<(mockSystemTime.nanoTime)
+  }.set(minTestsOk = 10)
+
+/** HELPERS */
+class MockSystemTime() extends SystemTime:
+  private var times: LazyList[Long] = LazyList.from(1).map(_.toLong)
+
+  def nanoTime: Long =
+    times match {
+      case t #:: ts => times = ts; t
+    }
+
+case class SmallInt(value: Int)
+
+object SmallInt {
+  given Arbitrary[SmallInt] = Arbitrary {
+    arbitrary[Int].map(n => SmallInt((n % 10).abs + 1))
+  }
+}