squashfuse_ll multi threaded mode is slower than both single threaded LL/high-level fuse

[isidentical]

For the same image (a compressed Python environment with -comp=lzo, -Xalgorithm=lzo1x_1), a mainly metadata heavy import procedure takes almost %25 more time when using the default options (multithreaded) compared to the single threaded mode of the squashfuse_ll.

Benchmarks

	first time	second time	third time
squashfuse	0m3.454s	0m2.646s	0m2.668s
squashfuse_ll	0m4.367s	0m2.454s	0m2.597s
squashfuse_ll -o clone_fd	0m4.418s	0m2.398s	0m2.407s
squashfuse_ll -s	0m3.389s	0m2.388s	0m2.366s

(tested on a freshly built version of squashfuse 0.5.0)

[DrDaveD]

Thanks for running the benchmark and reporting the results. Is the application single threaded? The default setting is for best performance with multithreaded applications.