While the QT implementation is sufficient for many use cases there are some problem with using it. One is that it is no longer free for all platform. They made a new highly optimized version that runs on many devices and it is NOT open source.
Making a custom drawn LCL implementation that is based on a Scene Graph and renders on TinyGL (fast software rendering), OpenGL, GL ES, Direct X and other hardware would have several advantages. Using less CPU, full custom rendering and portability would be the the most obvious one. OpenGL ES is supported by virtually all embedded devices as well as desktop MESA drivers.
Other advantages outside of the LCL context:
LCL + game/2D/3D integration. Having a scene graph as the layer below LCL means that other things can be drawn using 2D and 3D hardware acceleration. Game can run top level and contain an LCL UI as one of its nodes. And a game can run inside an LCL control. This could even be nested in various ways. Some of this can probably already be done but not in the same rendering pipeline. Rendering a non-GL gui on top of GL means CPU usage as well as unnecessary texture copying.
Hardware media integration. VAAPI, OpenMAX, and Vulcan are a few different way to use hardware together with zero copy rendering of video onto an OpenGL Scene. I am sure that DirectX have something similar. Implementing this inside a Scene Graph means that neither game nor LCL developers have to care what the underlying hardware is.
Raspberry Pi support. There are two ways to run applicatins on Raspberry Pi. Using Xorg either with or without OpenGL ES or using the Frambuffer with or without GL ES. The latter provides heavy advantages as Xorg consumes a lot of memory even when just running OpenGL ES applications. Wayland is also an alternative but it is not really in an usable state. I would suggest an implementation that can run directly on the framebuffer. This means that it will have to take car of all graphics such as rendering a desktop, window borders etc.
Android/Mobile support. I know that there are some attempt to implement LCL for OpenGL ES on Android. I would suggest that developing on desktop and then porting to android has several advantages. Lazarus itself currently only runs on desktop which means that developing for Android includes cross compilation and less than easy debugging. To implement a desktop solution perhaps on top of SDL and then port it to Android means that only platform specific bugs has to be hammered out on the Android platform. In short it will allow for a much faster development cycle. Also this means that all applications that works on desktop will then automatically work on Android, once bugs are hammered out.