Perhaps I made a mistake measuring this - I don't know.

I did a comparison of the code size for NetBSD (4.4/Mach) VM, UVM, and Linux's
VM system.  I measured object code size, as reported by nm -n.  I
included the swap subsystem in addition to the VM system, but I did not
include any filesystems, vnode stuff, device drivers, or i386-specific code.

Linux VM: 33K.
Mach/4.4 VM: 45K.  This VM system was derived from a microkernel implementation
  that tried to be much more general, so I can understand the slight increase
  in code size.
UVM: 144K.

Yowww!!!  Any reason for this?  Do we have large static tables in the code
segment (I ignored the data segment)?  Can I expect three times the 
functionality and performance?

:-)