I spoke to Kelly and Garima in Basemaps about this 2018 as well at UC in person.
Ideally, vector tile layers would have the option of having attributes. In other words, vector feature layers. Perhaps this would be a wise default. Best of both worlds. Right now it seems one must author maps with cumbersome duplicates of all layers; one with mapserver or featureserver invisible and then vector tile companion sublayer for each layer as the visible layer. Conceptually, I think we all simply want featureserver layers that are tiled and thus load super-fast like vector tile layers. But act like featureserver layers in terms of attributes and querying. (And if one needs different symbology, the layer should be able to do that. Of course it won't be tiled but can render on-demand; same as layers currently do. Perhaps it would even intelligently begin to cache on the client and/or server; like rasters can)
The big reason is performance. Even mapServer can hit servers hard. There are many, many bottlenecks but 1) customers may not know of them 2) if we do we sometimes can't change them. For example, you could export your SDE daily into a web mercator file gdb on the C drive of the webserver and load balance the webserver or put the gdb on a load balanced SAN. Or if it is pulling from SDE make sure it is posted and rec'd, compressed, etc. However sometimes some or all these things are out of our control. Users hitting a county parcel layer will hammer the server. No fear! Vector tiles cut through all of this blazingly efficiently. But, no attributes. Unless we do the featurelayer workaround. Which most of us don't have time for except for limited flagship apps. I understand they are .pbf and so forth but perhaps the API could do this 'twin layer' feature/vector thing under the hood and simply make it "look" like one layer. Maybe for like Arc 11.4 next year or something.
I too hope this can get traction soon. Many folks are finally drinking the hosted feature layer kool-aid now. This would be a big sellling item to get the rest of everyone on-board with data store. (Because, we are losing a lot of power with SQL Server Management Studio and the like for direct SDE access to do stuff and fix stuff; with the black-box Portal data store in PostgreSQL. So this would be a tradeoff that may make it worthwhile. Or perhaps we could set up replication to the hosted data store from SDE to retain best of both worlds)
