Well, I haven't actually played much with PDO, and it's been a while since I did any kind of testing against AdoDB. I would say that in general, AdoDB is better tested, since it's been around a while. PDO is brand spaking new so expect to be helping the PDO folks thrash out the bugs if you start using it. OTOH, it's probably got plenty of active development going on, so if you found a bug it'd likely get fixed fairly fast.
Btw, if you already support MSSQL and Oracle, you might as well work on PostgreSQL as well. It's got stored procs (OK, they're User defined functions, but other than the strange select procname() calling convention, they're the same thing.) and has had triggers and UDFs for years now.
I'd definitely start low level and see if the standard PHP mysql interface can handle return sets, as well as the mysqli interface, to see where the problem is. I know someone else on this list had a problem with mysqli and resource leakage, whereas I've never had a problem with resource leakage with the plain mysql interface, so that leads me to believe that the mysqli interface may not be "done" just yet.
After figuring out which interface supports returning result sets, see if you can put ADODB or PDO on top of them. PEAR:πb is the oldest of the abstraction models, and generally the least recommended nowadays. I'm guessing ADODB is likely to have the best combination of features and maturity for now. Look for PDO to pass it by in the next year or two, unless someone decides it's hopeless for some reason and development dies off. It's happened before, but I'm guessing there's too much momentum for PDO to just die off.