In my experience, LightSwitch (the HTML client without any 3rd party libraries... SL is EOL) is a good choice when these criteria are met:
1) You can build your database from scratch, or your database is normalized with proper keys. Many legacy DB's will not conform to LS's desires (unique keys), forcing you to write a WCF RIA service to adapt it. This kills a lot of the RADness.
2) You don't have huge datasets. LS basically attempts to read all of the entities at once and cache them, and queries it in-memory. Working around this requires an RIA service, again killing RADness.
3) Your application is mainly CRUD, or simple operations (approval, workflow type stuff).
4) Your application does not require a super-specified UI. If you have detail-oriented people who want everything just so, you are going to be in for some pain. Remember, it's made to work on everything from an iPhone's browser to IE11 on a 1080p display, so attempting to make it pixel-perfect everywhere isn't going to happen (nor should it). Not only that, but some "obvious" things are hard to do if LS doesn't do it for you. For example, binding a dropdown to a query result set for filtering requires some doing.
5) You are familiar with jQuery[Mobile] and fully comfortable reading and writing function-flavored javascript.
6) You are willing to poke and prod JavaScript, and basically operate half-blind. The documentation for LS HTML5 isn't all that great... you basically have to read from example code scattered here and there and figure out how to adapt it, and it isn't always clear how LS wants you to do things, and doing them the wrong way results in code that looks like it works, but will have subtle issues. The binding process isn't great: it works perfectly except when it doesn't, and when it doesn't, it's often not clear why. Intellisense rarely works well (understandably so... Javascript is a poor language for static analysis).
In general, unless it's pretty much perfect for LS, or you can let LS drive your requirements a little bit, or you are a freaking whiz at advanced javascript and like a little mystery in your APIs and living on the edge, I'd say pass on the HTML5 at the moment if you have other reasonable options.
As for the SilverLight version... shrug. I have no opinion technically, but in terms of long-term solutions SilverLight won't be viable much longer.
I will say LS's deployment is pretty awesome... drops right into IIS painlessly.