Debugging constraint solver performance issues is a real challenge; even a small mistake in a constraint can impact constraint solver performance . Common mistake people do which results in constrain solver performance degradation is to constraint the data payload to a predictable pattern to enable designers to debug RTL faster. When predictable data pattern is constrained for every transaction the number of bidirectional constraints solved increases exponentially resulting in constraint solver performance degradation. The problem would be magnified further when a scenario generator is used to randomize sequence of transactions.
The solution to this is if you would require a predictable data pattern to be used in the data payload override the random value with predictable pattern in post_randomize() method instead of going for a constraint, this would improve the runtime performance of your simulation.