Up side of the aproach is you are supposedly getting things done more quickly therefore increased productivity.
Apparent downside is that increased system resources needed for the application. The answer to that is buying memmory is cheaper than the developers time.
This approach apperntly works but still I don't think it is acceptable. My personel experience is that while you should use cluster of servers for your application you should also need powerful machines for your developers in order to get the developement time benefit. For example checking the hibernate forums lots of junior developers find out that they needed to extend "permanent heap size" the hard way. Again even with a powerful machine and hot deployment enabled server you will be restarting your machine a lot which is the faster the better. If you are using an api like hibernate which generates classes to do their trick and jvm with a permanent heap space, the more you restart your machine the more classes will be left in your memmory which means buy more RAM. The funny thing is as this is a developement problem it can be thoughted to be ignored.