I ran an insert workload and induced some replica set lag to create cache pressure. In about 5 minutes of running it generated about 2 billion lookaside table entries, for a LAS file size of about 1.2. GB (highly compressible data; would be much larger for more typical data).
In 3.6.3 I can see 10 k lookaside table removes every 10 seconds - is that the expected effect of this change? At that rate it will take about 3 weeks to get rid of the lookaside table entries that took 5 minutes to generate, so I'm wondering about the efficacy of this; maybe it needs some tuning?