If you have them often enough, I think the transience factor isn't a problem. The tempo of Facebook hackathons seems pretty good to me - somewhere around 8-10 weeks or so between the last three.
Doing it once or twice a year in isolation is probably too infrequent to get lasting value - the positive value might have a total lifetime of about 4 months, so I'd suggest three or four times a year is a minimum - unless you have some other mechanisms that try to solve the same issues that hackathons do.
Doing it once or twice a year in isolation is probably too infrequent to get lasting value - the positive value might have a total lifetime of about 4 months, so I'd suggest three or four times a year is a minimum - unless you have some other mechanisms that try to solve the same issues that hackathons do.