On the objective stuff, per Apple's 7500/8500 hardware devnotes, Page 20:
The video subsystem is implemented by the following ICs:
■ Chaos, a custom IC that provides data bus buffering between the video subsystem and the processor bus
■ Control, a custom IC that provides addressing and control for the video subsystem
■ RaDACal, a high-performance digital-to-analog converter (DAC) used for the video stream to the monitor
■ Sixty6, an RGB-to-YUV converter and convolver for the second video output stream (on the Power Macintosh 8500 computer only)
■ a 7187 digital video encoder (DENC) for the second video output stream (on the Power Macintosh 8500 computer only)
■ an 8758 analog-to-digital converter (ADC) for the video input stream
■ a 7196 digital video decoder and scaler IC (DESC) for the video input stream
■ Plan B, a DMA controller for video input data from the 7196 DESC IC
Elsewhere in that section only very limited forms of hardware acceleration are hinted at:
The RaDACal IC also ... supports the hardware cursor.
The video input feature is implemented by the 8758 video ADC IC ... video data can be stored either in the video display buffer (VRAM) or in an offscreen pixel map in main memory. Video data transfers are DMA transfers and are controlled by a DBDMA engine in the Plan B IC.
When video input data is sent to the display buffer, it can be clipped with a 1-bit-per-pixel clip mask using the DBDMA read channel. ...
When video input data is stored in a pixel map in main memory, software can perform any required clipping and blending by using the CopyBits routine in QuickDraw when moving the pixel map to a visible region of the display.
So it appears as if there was a fast hardware path for dealing with video input (in the AV sense), including depositing it at an appropriate on- or off-screen destination and clipping it if it is sent on-screen. It also seems that any graphics, once in memory, are manipulated in software using the standard Apple calls.
I therefore think the answer to your question is that no hardware acceleration was provided for ordinary 2d graphics.
On the subjective, software rendering was indeed more than good enough for the typical resolutions and usages of the mid-'90s, on both the PC and the Mac — to the point that many of the pre-3dfx/PowerVR 3d accelerators were derided for doing a worse job than software rendering, e.g. from 2003:
It may seem hard to believe now, but ten years ago, S3 was the market leader for 2D-only GPUs. But as the 3D revolution gathered steam in 1996, it became clear that S3’s ViRGE GPU wasn’t going to allow S3 to maintain its dominant market position. In fact, the ViRGE became known as a “3D decelerator”.
1996 is the year after the introduction of the 8500.
If you want easy YouTube points of reference then check out the Apple Pippin — it has a PowerPC 601, substantially slower than the 604 in the 8500 — and no video or audio acceleration whatsoever but manages to throw around mild 3d scenes at TV resolutions; suffice to say that at desktop resolutions a 604 is more than a match for the more sedate world of that era's desktop 2d.