I'd look at ERC+ or ERC#. The straight ERC (and other stats) can be misleading based on the season. Pitchers from low-scoring seasons will have better ERC, but their ERC+ would be compared to other pitchers from that season and ERC# would be normalized across the eras. For example:
1908 Mathewson ERC 1.38 ERC+ 182 ERC# 1.74
2008 Martinez ERC 1.39 ERC+ 353 ERC# 1.18
That said, there are those (zubinsum, I believe, among others big on spreadsheets and the math behind the game) who wonder why anyone looks at stats not used by the engine. I find the ERC+/# numbers useful as a summary of pitchers' abilities. The straight (ERC) numbers are a good basis in a single-season progressive, when all the players being considered would normalize the same way. But in Open or most Theme leagues, use normalized stats. + figures relate to the player's league and season; an ERC+ of 100 means a pitcher was exactly league average. # stats compare all eras.
Finally, it's a good idea to lok at the $/IP as a check. If you think Pitcher A and Pitcher B are similar, and B has a significantly higher $/IP, that means there's something about Pitcher B that the sim engine likes.