Simulating continuous time semi-Markov state machine and changing transition probability on the fly

Question

The problem that I'm trying to solve (well, I think that I almost did, but need a review from someone more experienced) is about changing probability of the transition for semi-Markov state machine during simulation.

Let's assume that M=<S,T> defines machine with S states and T probabilistic transitions between states. Elements of T are triplets (s_out, s_in, lambda). So the probability of moving from state s_out to s_in is exponential and defined as exp(-lambda*t).

To simulate it I get random value from [0;1), transform it to exponential and save as next event. What happens is that after saving it but before invoking another event may occur (so it is not a pure Markov state machine) and change the lambda. So the event time should be recalculated.

The problem is that it cannot be just recalculated as such changes may occur quite frequently constantly shifting the event time to the future.

What I think is correct is to adjust recorded event time on ratio time_to_event_from_now * lambda_old / lambda_new. Does it sounds right to you? Maybe the problem description recalls the paper/book where this problem solved?