20K rules is not excessive, and this number alone should not cause out-of-memory exceptions. Mainly, this is due to rules not being well designed, and in combination with heavy fact loads.
Frequently, a large number of rules results from putting all the decision data into rule conditions, even when there are many rules with identical structure in condition and consequence. Offloading the varying constants into static facts may reduce the number of rules by magnitudes.
Not the number of rules but the maintenance issue is addressed by using decision tables or some other technique of generating rules from a handier source.
-W