3 Way Compression, Masked Compression

[github]

tested the Multithreaded Precomp V 0.48,
when I have setup this

[External compressor : precomp]
header = 0
packcmd = cls-precomp_x64 -cn -intense -t-j3 -o$$arcpackedfile$$.tmp $$arcdatafile$$.tmp

The multi thread works, but it will call the cls-precomp.exe instead of cls-precomp_x64.exe

The compressing time reduced a little bit, and the result is 4% smaller?
Which doesn't seem to be right, multi thread should only reduce time not result size.
When try to unpack the result, it says
ERROR: archive structure corrupted (decompression of control block failed)Press any key to continue . . .

Reproduce:

mask=precomp+srep64+xz

[External compressor : precomp]
header = 0
packcmd = cls-precomp -cn {option} -o$$arcpackedfile$$.tmp $$arcdatafile$$.tmp

[External compressor:srep64]
; -mem25% -mem1024mb
header = 0
packcmd = srep64 {options} $$arcdatafile$$.tmp $$arcpackedfile$$.tmp
unpackcmd = srep64 {options} -d -s - - <stdin> <stdout>

[External compressor:xz]
header = 0
packcmd = xz a -txz -an -mcrc=0 -m1=lzma2:d100m:fb=273:mf=bt4:mc=1000000:lc=4:lp=0 -mmt=14 -mx9 -si -so <stdin> <stdout>
unpackcmd = dec x -txz -an -y -si -so <stdin> <stdout>