Normally I'll just bookmark papers, but I've found one that deserves better treatment: Scalable Lock-free Dynamic Memory Allocation. It was presented at PLDI04 and is everything a paper should be: practical, clear and with great results.
The author presents a malloc implementation which is lock-free (so you can call it from signal handlers, or kill threads and not cause deadlock etc) and it's faster (on SMP boxes) than the other mainstream concurrent allocators (Hoard and PTMalloc specifically).
For anyone at Imperial, you should come to a talk by Tim Harris on this subject on March 9th (details to be announced).
And everyone at (or near) Imperial should come to the charity ball in the Great Hall on the 26th of this month. Tickets are £10 and every penny goes to charity.