I'm going to give you one more thing to think about if ultimate apparent sharpness is your goal: how well the actual details fit the pixel matrix you are using to represent them.
When you shoot at the finished size, you are stuck with the pixels the capture device gives you at that resolution. You can run sharpening and deconvolution algorithms on that data until the cows come home, but you can't change the starting data.
If you shoot at double the final resolution and do an overall sharpening first, you can change the crop before resizing. A one-pixel horizontal or vertical change in the position of the crop doesn't sound like much, but it can have a huge impact on how well the most important details (say, the maker's name or the numbers on the watch face) align with the final-sized image's pixels.
You can automate the process (crop then resize) in most image editors, giving you a one-click action to create four reduced-sized images that you can select from to create the final web image. It should be apparent that one of the images in the 4-image set is the best starting point for final output sharpening, depending on the specific detail you want to concentrate on most. It wouldn't make nearly as much of a difference with organic shapes, but it can have an enormous impact when you're dealing with fine, high-contrast artificial detail.