In response to the various comments & answers about measuring rather than guessing, and also to the comments regarding the count of rows where col2 is NULL
vs col2
having a value of some kind, I ran 4 tests on a test table:
| Query 1 | Query 2
col2 almost all NULL | 5 min 27.96 sec | 4 min 20.38 sec
col2 almost all has values | 34 min 31.64 sec | 32 min 31.54 sec
Conclusion: Query 2 is marginally faster in both cases.