I still like channels because they may be a net reduction in the number of concurrency primitives in use, which complicates quantification in the paper - their taxonomy is great, though. Channels have some sharp corners.