you are really measuring strange things on your sytsem

Yeah :( It seems to the result of a conflict between 64-bit perl 5.10, and the MS runtime. I get a much smaller differential with 32-bit 5.8.8, but it's still not great:

C:\test>\perl32\bin\perl -MTime::HiRes=time -we"my@a=0..1e6;my$t=time;
+my@bla=@a; tr[0][]d for @bla;print time-$t; <>"
0.638000011444092

C:\test>\perl32\bin\perl -MTime::HiRes=time -we"my@a=0..1e6;my$t=time;
+my@bla=map{tr/h//d;$_}map{$_}@a;print time-$t; <>"
9.05199980735779
[download]

But every test I tried using malloc()/free() from the 64-bit MS runtime using C seem to fly. Course, it'd be easier to track down if you could get some feedback on these issues, but that's never forthcoming.