rendering a model with +64k requires mesh vertex/index buffer support above 16bit .
back in the days, hardware had the 16bit limit ,nowadays we have hardware that can handle more ,like 32bit , but acknex is old tech and the limit in it remains at 16bit for mesh buffers.
when you have your model "chunks" (nodes) under this limit , in your modeling application , and need to stay within the limits of acknex engine , you have to remember that everything exported in the mesh file ,will be the mesh when imported , the keyword here is "mesh" , even though you may have submeshes (nodes) which make up the whole "mesh" , the final mesh consist's of these submeshes as a single mesh buffer when imported ,i believe .
therefore if in total ,your submeshes , together , consists of more than the limit at which the engine is at (16 bit in this case) , you have options of either two things ,
A) code the file loader yourself ,and seperate the nodes to become each a seperate model mesh .
B) export each node as a seperate mesh on theire own.
the principle is quite simple , in general , each chunk/mesh may consist of more than one submesh with different materials/shaders but , as long as each chunk/mesh consists entirely as a "whole" in total when added together ,as a single buffer under the limit of what the tech supports.
the problem here becomes , what the model file loader does at loading time with nodes , if it would have been able to split buffers based on the set limits when reading the file , it would have had so much more complications in regards to what the engine is and does , *totaly opinion based reply by the way.