Unless you’ve a good reason, I would always choose Open as Smart Object. Good reasons might be you’re running out of hard drive space, or you have an earlier version of Photoshop than your version of Lightroom.

OK, now why? I tend to assume there’s a fair chance you’ll want to fine tune the raw conversion at some stage in the future. For example, a new version of Camera Raw may have better noise reduction and you may want to rework the picture. Alternatively, you may have overlooked some dust spots and prefer to correct them at the raw level rather than with a retouching layer, or you might notice a lens aberration of some kind. Smart Objects let you do this.

Secondly smart objects allow you to adopt some very effective ways to work. For example, filters are applied as smart filters which unlike regular filters are non-destructive, so you can fine tune them long afterwards – so my workflow to Silver Efex is LR, Smart Object in Photoshop, then Silver Efex. Cropping or transform operations can be done non-destructively with smart objects too. If I do need to do some pixel level cloning, I’ll simply copy from the smart object layer into a regular pixel layer. So there’s no real downside.