The maximum is 8309 ($2075) nibbles for track 0. Well, according to a little experiment I did. ; - )
The Disk II uses Constant Angular Velocity standardized by Shugart at 300 RPM. The earlier 8-inch drives were 360 RPM, and since the physical media was the same it's likely that 300 RPM was chosen to increase data density on the smaller disks, and perhaps reduce controller throughput for use on early computers that were less powerful and cheaper. (But I can't find a reference for this.)
Disk II spindle speed can be manually adjusted on Apple's analog controller board between approximately 190 and 320 RPM. Slower speed results in higher data density, but this will be limited by physical parameters such as magnetic (iron oxide) particle density and read/write head characteristics - e.g. weaker magnetically induced current in the read head at low speed.
Track capacity is affected by its contents: since MFM clock pulses aren't used, the data written must be carefully chosen in order to synchronize with and be reliably read by the Disk II controller. This includes restrictions on consecutive zeros, and nibbles with trailing zeroes called self-sync bytes.
However in this experiment I didn't use sync bytes, instead relying on the controller sequencer's natural tendency to quickly sync to certain patterns. A simple nibble count measures the maximum theoretical capacity of the track, ignoring the required overheads of practical data storage. Here is the nibble count code:
ORG $8000
LDA #$30 ;WRITE LEN
STA $01
LDX #$50 ;SLOT * $10
LDA $C08D,X ;LOAD WP
LDA $C08E,X ;READ WP
BPL NOTWP
RTS
NOTWP LDA #$D5 ;1ST NIBBLE
STA $C08F,X ;WRITE MODE
CMP $C08C,X ;4 - SHIFT
BIT $FF ;3
LDY #$00 ;2
LOOP1 DEC $FF ;5
EOR #$7F ;2 - D5^AA
LOOP2 DEC $FF ;5
CMP $FFFF ;4
NOP ;2
STA $C08D,X ;5 - LOAD
CMP $C08C,X ;4 - SHIFT
DEY ;2
BNE LOOP1 ;3 OR 2
DEC $01 ;5
BNE LOOP2 ;3 OR 2
DEC $FF ;5
DEC $FF ;5
MARK LDA #$FF ;2 - MARK
STA $C08D,X ;5 - LOAD
CMP $C08C,X ;4 - SHIFT
JSR RTS ;12
LDA #$20 ;2 - ADDR
STA STORE+2 ;4
STA COUNT+2 ;4
LDY $01 ;3
STY $00 ;3
LDA $C08E,X ;READ MODE
READ LDA $C08C,X
BPL READ
STORE STA $2000,Y
INY
BNE READ
INC STORE+2
BPL READ ;TO $8000
STA $C088,X ;MOTOR OFF
LDX #$00
COUNT LDA $2000,X
CMP MARK+1
BNE NEXT
LDA COUNT+2 ;PRINT ADDR
JSR $FDDA
TXA
JSR $FDDA
LDA #$A0
JSR $FDED
LDA $01 ;1ST MARK?
BEQ COPY
TXA ;NO
SEC ;CALC LEN
SBC $00
PHA
JSR $FE80 ;SETINV
LDA COUNT+2
SBC $01
JSR $FDDA ;PRINT LEN
PLA
JSR $FDDA
JSR $FE84 ;SETNORM
LDA #$8D
JSR $FDED
COPY LDA COUNT+2
STA $01
STX $00
NEXT INX
BNE COUNT
INC COUNT+2
BPL COUNT ;TO $8000?
RTS RTS ;YES, DONE
Assembled at $8000, there are some defaults that can be optionally adjusted:
- 8001: The number of nibbles to write to the track in units of $100 (Default $30 = $3000 nibbles)
- 8005: The drive slot number times $10 (Default $50 - safer than $60!)
- 8010: The first nibble value; this gives a pair sequence via XOR with $7F (Default D5/AA)
- 801E: The XOR value; bit 8 should be 0 (Default $7F)
- 8037: The nibble count mark value (Default $FF)
- 8042: The page to start storing track data when read back and nibble counted (Default $20)
The routine assumes the drive is on and up-to-speed, and that the head is already positioned on the desired track. The defaults as given will write $3000 nibbles of "D5AA" followed by one "FF" mark to the selected drive in slot 5. The track will then be read into memory at $2000-$7FFF and the data scanned for the mark value. Mark locations will be printed in normal text, and length between marks in inverse.
Here is an image of a sample run which uses boot 0 to seek track 0 of drive 1 in slot 6 in AppleWin:

Adjusting a real drive's speed (measured with Copy II Plus 8.4) I obtained the following results:
Speed (ms) Speed (RPM) Max length Length (Hex)
188 319 6015 177F
200 300 6400 1900
210 286 6715 1A3B
220 273 7045 1B85
230 261 7360 1CC0
240 250 7680 1E00
250 240 8004 1F44
260 231 8309 2075
270 222 - -
280 214 - -
317 189 - -
At 270ms and above the sequencer would not auto-sync to simple nibble patterns, though Copy II Plus could still measure drive speed. Its exclusive use of selected sync byte values would likely explain this.
Tracks written at one speed can be read at another speed that is +/-10ms. I didn't test how far this tolerance extends. Roland Gustafsson (who wrote the copy protection for many Brøderbund titles) says he tried slowing the drive speed but "never used that technique due to compatibility problems. In fact, [he] wrote a speed calibrator that was used at Broderbund to keep the disk drives in spec".
When speculating about an alternative Constant Linear Velocity capacity you would need to specify the exact hardware being used, since that big a change would probably extend from the controlling software to the magnetic media - and everything in between. There is a good list of existing floppy disk formats on Wikipedia.