[Some commercial service I'm redacting mention of] has clients that employ the exact failover scenario you described.
The difference is that the system is completely automated. It monitors your servers and modifies the dns records on the fly. It reacts quite a bit faster than an admin getting a page, crawling out of bed, putting on the coffee pot, and changing the records manually. It is also immune to the "just woke up, fat-fingered a record" effect :D
In real life observations, the current client behaviour is that they do flood over the live server pretty quickly.
Despite the fact that there may be some disruption, if the only other option is to be completely down, then being available for the great majority of clients si a pretty good option.
The 48 hours mentioned above does not exist for the contemplated usage. That is for dns propogation time which is related to changing the registered name servers at the registrar level. This is not the same as changing records at an unchanged name server.