The following links are about "deeper magic" of DTrace internals,
in particluar, challenges of faithfully tracing multi-threaded
processes.

It is hard to debug a multi-threaded process with a traditional 
debugger, because such debuggers eliminate concurrency (or, "serialize")
threaded code -- making it impossible for some bugs to be reproduced
under the debugger (cf. "Heisenbug"). 

http://dtrace.org/blogs/bmc/2011/03/09/when-magic-collides/
  (note discussion of loss-less tracing of threads)

http://dtrace.org/resources/ahl/img0.html  (skip SPARC slides, note
  the discussion of sync/async signals, amd64 RIP-relative addressing)