"Fixed Time, Tiered Memory, and Superlinear Speedup,"
Proceedings of the Fifth Distributed Memory Computing Conference (DMCC5),
Much has been written about superlinear speedup, where a collection of N processors seems to run more than N times faster than a single processor. This was the first paper to point out how absurd it is that this should surprise anyone, particularly if one thinks more carefully about what "speedup" means.
It was also, other than a letter to Communications of the ACM, the first place the concept of fixed-time benchmarking was spelled out.