Smart Objects offer some wonderful flexibility in Photoshop; they allow endless transformations without a cumulative loss of data, because each transformation is a fresh start from the original data, whether that data is raster or vector. Why, you can even open a Camera Raw file as a Smart Object, which allows you to go back to the Camera Raw data for color correction, noise reduction, or straightening—it’s the ultimate in non-destructive editing.
But there are two scenarios in which Smart Objects act a bit dumb.
Rendering Vector Content
If you create a Shape layer or text layer in Photoshop, then save the file as a Photoshop PDF, the text or shape will render as vector when the PDF is placed into InDesign or Illustrator. If you save the file as a PSD, the vector content will display and render as pixels, at the resolution of the underlying image.
However, vector Smart Objects will render as pixels even if you save the file as a Photoshop PDF. My take is that the vector Smart Object data is used as a resource to generate pixels, and not directly available to other applications. It’s not a showstopper, but something to consider; in some cases, you might be better off pasting (rather than placing) vector content, manipulating the vectors, then converting to a Smart Object.
If you assign a blending mode to a layer, then convert that layer to a Smart Object, the blending mode is honored within that image.
However, if you place an image with a layer using a blending mode (thus making it a Smart Object on the way in the door), blending modes are NOT honored. I’ve used shadows with the Multiply blend mode as an example, but this applies to all blending modes.