I understand your priority. I heard a lot of great features are coming next year.
I did a simple benchmark on a single node MemSQL instance running cluster-in-a-box (master aggregator + leaf).
Binary UUID: BINARY(16) with regular UUID version 4
Ordered UUID: BINARY(16) with ordered UUID version 1
Comb UUID: BINARY(16) with comb UUID version 4
Textual UUID: CHAR(36) with regular UUID version 4
Each run selects a random UUID from the dataset 10,000 times with index and calculates the average time.