Thanks everyone for the assistance.
Regarding the questions about the server, this was a situation where I
take a single database backup file, restore it on machine A and machine
B, and the execution plans are different for the same query. Assume A
is bad and B is good. If I take a different backup file (same basic DB
just from a different date), and put it on A and B, I get the same
results. Meaning it's something about the MACHINE, and not something
about the database.

So, it turns out that I overlooked the most obvious answer to the
question, and that is that the bad machine actually had only one CPU!
So parallelism is not used. I never even checked for this because I
thought I already knew that it had two CPUs. My bad. They must have
been just barely sneaking in under the timeout value earlier, but the
extra month of data pushed it over the brink.

I did find the OPTION (maxdop 2) hint, and also OPTION (hash join),
which would help if I could actually use them!
Thanks again everyone.
Christine