It depends on you requirements, but given the back-of-the-envelope calculations you have a quick way to compare them without building them.

Now you have a framework for asking yourself other design questions and comparing different design variations:

Does it make sense to cache single thumbnail images?

Should you cache a whole set of images in one entry?

Does it make sense to precompute the thumbnails?

To make these estimates realistic you'll have to know the performance of your services. If there is an unknown variable then perhaps you could rapidly prototype just that part to settle the question. To know if caching is a good design alternative, for example, you'll have to know how long it takes to write into your cache.