I've always found the repeated application of rules to simplify patterns in expressions until the expression stops changing to be pretty powerful when just playing around, especially with small problems. However, in my experience in bioinformatics, that flexible approach is so slow in the face of actual data as to be totally useless for use on real problems in my line of work.

I think an option to use a faster, less powerful, linear pattern matching algorithm more like in ML/Haskell would help a lot in this regard.