diff --git a/lib/system/threads.nim b/lib/system/threads.nim
index 49b13576ce..bb03b6d0ed 100644
--- a/lib/system/threads.nim
+++ b/lib/system/threads.nim
@@ -115,10 +115,6 @@ when defined(windows):
   proc setThreadAffinityMask(hThread: SysThread, dwThreadAffinityMask: uint) {.
     importc: "SetThreadAffinityMask", stdcall, header: "<windows.h>".}
 
-  proc getThreadId*(): int =
-    ## get the ID of the currently running thread.
-    result = int(getCurrentThreadId())
-
 elif defined(genode):
   const
     GenodeHeader = "genode_cpp/threads.h"
@@ -249,48 +245,6 @@ else:
   proc setAffinity(thread: SysThread; setsize: csize; s: var CpuSet) {.
     importc: "pthread_setaffinity_np", header: pthreadh.}
 
-  when defined(linux):
-    proc syscall(arg: clong): clong {.varargs, importc: "syscall", header: "<unistd.h>".}
-    var NR_gettid {.importc: "__NR_gettid", header: "<sys/syscall.h>".}: int
-
-    #type Pid {.importc: "pid_t", header: "<sys/types.h>".} = distinct int
-    #proc gettid(): Pid {.importc, header: "<sys/types.h>".}
-
-    proc getThreadId*(): int =
-      ## get the ID of the currently running thread.
-      result = int(syscall(NR_gettid))
-  elif defined(dragonfly):
-    proc lwp_gettid(): int32 {.importc, header: "unistd.h".}
-
-    proc getThreadId*(): int =
-      result = int(lwp_gettid())
-  elif defined(openbsd):
-    proc getthrid(): int32 {.importc: "getthrid", header: "<unistd.h>".}
-
-    proc getThreadId*(): int =
-      result = int(getthrid())
-  elif defined(netbsd):
-    proc lwp_self(): int32 {.importc: "_lwp_self", header: "<lwp.h>".}
-
-    proc getThreadId*(): int =
-      result = int(lwp_self())
-  elif defined(macosx) or defined(freebsd):
-    proc pthread_threadid_np(y: pointer; x: var uint64): cint {.importc, header: "pthread.h".}
-
-    proc getThreadId*(): int =
-      ## get the ID of the currently running thread.
-      var x: uint64
-      result = pthread_threadid_np(nil, x)
-      result = int(x)
-  elif defined(solaris):
-    # just a guess really:
-    type thread_t {.importc: "thread_t", header: "<thread.h>".} = distinct int
-    proc thr_self(): thread_t {.importc, header: "<thread.h>".}
-
-    proc getThreadId*(): int =
-      ## get the ID of the currently running thread.
-      result = int(thr_self())
-
 const
   emulatedThreadVars = compileOption("tlsEmulation")
 
@@ -665,3 +619,82 @@ when useStackMaskHack:
     var mainThread: Thread[pointer]
     createThread(mainThread, tp)
     joinThread(mainThread)
+
+## we need to cache current threadId to not perform syscall all the time
+var threadId {.threadvar.}: int
+
+when defined(windows):
+  proc getThreadId*(): int =
+    ## get the ID of the currently running thread.
+    if threadId == 0:
+      threadId = int(getCurrentThreadId())
+    result = threadId
+
+elif defined(linux):
+  proc syscall(arg: clong): clong {.varargs, importc: "syscall", header: "<unistd.h>".}
+  var NR_gettid {.importc: "__NR_gettid", header: "<sys/syscall.h>".}: int
+
+  proc getThreadId*(): int =
+    ## get the ID of the currently running thread.
+    if threadId == 0:
+      threadId = int(syscall(NR_gettid))
+    result = threadId
+
+elif defined(dragonfly):
+  proc lwp_gettid(): int32 {.importc, header: "unistd.h".}
+
+  proc getThreadId*(): int =
+    ## get the ID of the currently running thread.
+    if threadId == 0:
+      threadId = int(lwp_gettid())
+    result = threadId
+
+elif defined(openbsd):
+  proc getthrid(): int32 {.importc: "getthrid", header: "<unistd.h>".}
+
+  proc getThreadId*(): int =
+    ## get the ID of the currently running thread.
+    if threadId == 0:
+      threadId = int(getthrid())
+    result = threadId
+
+elif defined(netbsd):
+  proc lwp_self(): int32 {.importc: "_lwp_self", header: "<lwp.h>".}
+
+  proc getThreadId*(): int =
+    ## get the ID of the currently running thread.
+    if threadId == 0:
+      threadId = int(lwp_self())
+    result = threadId
+
+elif defined(freebsd):
+  proc syscall(arg: cint, arg0: ptr cint): cint {.varargs, importc: "syscall", header: "<unistd.h>".}
+  var SYS_thr_self {.importc:"SYS_thr_self", header:"<sys/syscall.h>"}: cint
+
+  proc getThreadId*(): int =
+    ## get the ID of the currently running thread.
+    var tid = 0.cint
+    if threadId == 0:
+      discard syscall(SYS_thr_self, addr tid)
+      threadId = tid
+    result = threadId
+
+elif defined(macosx):
+  proc syscall(arg: cint): cint {.varargs, importc: "syscall", header: "<unistd.h>".}
+  var SYS_thread_selfid {.importc:"SYS_thread_selfid", header:"<sys/syscall.h>".}: cint
+
+  proc getThreadId*(): int =
+    ## get the ID of the currently running thread.
+    if threadId == 0:
+      threadId = int(syscall(SYS_thread_selfid))
+    result = threadId
+
+elif defined(solaris):
+  type thread_t {.importc: "thread_t", header: "<thread.h>".} = distinct int
+  proc thr_self(): thread_t {.importc, header: "<thread.h>".}
+
+  proc getThreadId*(): int =
+    ## get the ID of the currently running thread.
+    if threadId == 0:
+      threadId = int(thr_self())
+    result = threadId