A few comments, hopefully of some use:

Q: It appears that mid-side node temperatures in SHELL93 are ignored.

A: I think that there is confusion on the BF vs. BFE commands. I do not know if this is by design or not, but

• BF applies temp load to nodes. Only temps for corner nodes may be input, and midside nodes are ignored. As a result, midside nodes assume linear variation between corner nodes. There is no assumed through-plane thermal gradient with this method. I attached a simple file tshell93.inp which shows this -- BFUNIF and midside nodes are set at 100, but the EPLOT does not show any nodes at 100.
  
• BFE applies temp load to "corners" (as defined in online help for SHELL93). There are 8 corners describing bottom and top corners. There can be a through-plane thermal gradient with this method.

I got burned once a while ago by this difference in the past since, to me, it is very confusing. Like I mentioned, I don't know if this is by design or not. Also, reading the documentation *carefully* is helpful since "STLOC" for BFE has different behavior, depending on how many inputs you have.

A: LDREAD uses BF, I believe, but I don't think it matters. The reason is because SHELL57 conducts in-plane only. Even if one tries to put convection on top & bottom surfaces, no through-plane conduction is considered. I attached a simple example, tshell57.inp which demonstrates this. In it, we have three thermal resistances in series -- convection on top, conduction through-plane, and convection on bottom. One may expect three thermal resistances in series, but there are only two -- conduction through-plane is not considered. (Look at NMISC,3 and NMISC,9 and you'll find they are the same) Hence, only nodal temperatures are calculated and transferred to structural shells (either bilinear or serendipity elements).

I believe that other FEA codes have through-plane integration points for their thermal shells where nodal temperatures are stored and can be transferred to structural elements -- ANSYS does not have this. (Nor are there any higher-order shells, as you noted)

Also, please keep in mind that all structural SHELLs have plane stress behavior. The through-plane thermal gradient will not cause any through-plane stress (there also is no way to constrain the "bottom" or "top" of a structural shell since constraints are at nodes (midsurfaces) only). The through-plane thermal gradient can cause out-of-plane bending, though.

A: For reasons above (hopefully clear from my rambling), no, I don't think you have to worry about midside node temperature input with SHELL93 when using LDREAD from a thermal analysis.

I think that if you do a /PBF,TEMP,,1 and the temperatures look fine from a LDREAD from SHELL57, you should be OK. One can always do a simple test case to confirm this.

A: Yes, it is. However, NMISC for temperatures at top and bottom surfaces are only for faces with convective input. I don't think it will give you what you are looking for since, as noted above, SHELL57 does not conduct in-plane. It does not store through-thickness temperatures (at integration points) like some other FEA codes do, so you can't get these thermal through-thickness effects into your SHELL93 model.