From a protocol viewpoint, I think there might be an issue of practicality. This is one man's opinion and if something has changed regarding Access network behavior, it's new to me.
Regardless of how you attempt to encapsulate network traffic, and if we are talking about "real" Access, there is a layer - no matter HOW deeply you encapsulate it - that must be a channel suitable for SMB network exchanges.
IF the app is a stand-alone (i.e. not split) app file, it might be POSSIBLE to create an environment suitable to run Access. If the connection is purely internal, network continuity is trivial via the phone's network stack's LOOPBACK connection at address 127.0.0.1, which every internal network active member will have.
IF the app is split with the FE on the phone and some other DB engine as the BE, all of the protocols I know require a formal connection environment because in the TCP vs. UDP dichotomy of network protocol styles, Access stays on the TCP side of the environment. Phones can and will sometimes be in a "no bars" area, which means the phone has experienced a "network drop" - which will pretty much kill any remotely linked session. The FE would crash. The BE would be subject to the ills of a crashed FE that might have had a pending operation.
IF the app is split and the BE is on the phone, the same rules apply as having the FE on the phone, because a network drop is symmetrical. A drop on one end corresponds to a drop on the other end.
If you have a situation with a UDP-class listener that is web-capable on the phone, AND you have some kind of app that you can launch that will translate UDP traffic into TCP traffic, you might get away with it until something times out and drops connections. A dropped connection will always be really bad for Access-based apps. "Bad" in this case would be a matter of kind, not a matter of degree.
Even for CITRIX connections which isolates the Access SMB traffic, if the CITRIX connection drops, you still have a hung instance that will eventually have to be shut down. In essence, that deferred shutdown delays the inevitable problem of a potentially incomplete update from a lost thread.
Let me say I am skeptical of making this work smoothly and reliably.