The problem with Howards BABIP (people should just call it H% like Shandler does since its so much easier~) was that it was too 2nd half heavy. He had a 42% H% in the 2nd half, nobody sustains that, nobody.
H% is a tool that really relies on past data to be useful. Its also not a predictive stat, its a stat that you use to look for the real ability of a player from the previous year. You don't use H% trends to decide if a player is improving or getting worse etc, its useful for figuring out if a jump in ability the previous year was a fluke.
A good example is Gary Mathews last year. His H% went from 28 to 35 and his AVG jumped from .255 to .313. You can pretty much be assured that the .313 is just a fluke. His normal H% is around 30-31 though so most likely the .255 was a fluke too. His real ability both years was probably in the .270-.275 range.
However last year his contact rate actually jumped 3% over his career norms. Usually contact rate predicts future success so I expect him to hit higher this year than that .270-.275 range, probably closer to .280-.285 assuming he doesn't lose AVG from the park switch.
Also, someone should determine some ratio for "pitcher's tendency to not have his best stuff." It would have to be based on the pitcher's own admission, though, at post-defeat press conferences.
Actually Baseball HQ has developed a stat called qERA. It rates each start by a pitcher from 1-5 based on how good it was. Then uses a ratio of good starts vs bad starts to define ERA. Generally speaking what they have found is pitchers who are consistently mediocre almost never improve but those who are dominant at times and then disasters at other times do end up becoming better pitchers a lot more.
When looking at whether or not qERA was predictive the results were pretty good. From 2001+2002 the system predicted big improvements in 27 starters and 22 of them did improve. It predicted dips in production for 62 pitchers and 46 of them declined.