I don't have a manual handy for that particular drive, so I am going to throw out some WAGs.

1. Is the VFD hitting a current limit and set up to reduce output to avoid faulting?
2. Wildest of the WAGS: Is the speed reference restricted only to the Modbus command? I have worked with some drives that by default will SUM all the activated references and you have to drill in and defeat that capability. One drive in particular had me scratching my head for an hour over this, it was picking up noise from a 0-10v terminal that had nothing wired to it and it was jumping around. This is a stretch, but something to look for.
3. Is VFD set up for PI(D) control and you are writing a process value or setpoint instead of a speed command? This could account for the drive variation if it thinks there is a setpoint and a process variable and it thinks it is in charge of adjusting to make them match. This should not result in hunting unless the SP or PV was changing though.
4. Could there be a timing issue where the Modbus timeout in the drive is just slightly shorter than the time between messages from the PLC? The drive is stopping because it thinks the comms have failed, but then it gets fresh data and picks back up.

All these are unlikely except perhaps the first one, and then you wouldn't expect it to overshoot the command reference by 5 Hz (maybe 0.5 Hz)...

its no where near its current limit.yes I set up all the references to take reading from the modbus signal .I do believe this drive does sum up the references .to me it seems like the drive wants to get a reference constantly like for an example a analog signal would be continues .with modbus I don't think I can achieve this because I use a Drum sequence to send the send and receives. which in turn every 500 ms it goes to next step .I have 4 send/ receives so it gets its reference every 2 seconds .and the drive is not stopping and stating up,so every 2 sec is good for the command but some reason not the reference .and the danfoss tech support is not of any help I keep getting a run around which is typical for those companies .thank you ADC for the best tech support in the industry.

I haven't used a Drum for Message sequencing, but I typically space them with a timer and set it for 50ms or so. I occasionally run into a situation where too fast causes hiccups on older stuff, but 50ms seems to be plenty slow enough and I don't know of a good reason to go slower than that.

All drives will require that you read or write something in order for the drive to "know" the comm channel is still working. They have all some sort of time period that will cause an alarm or a fault (usually configurable) if no comm port activity occurs.

Some of the VFDs I have worked with have a default comm loss time setting of 5 seconds and I will turn that up to 10 seconds (water pumps are my field and it is safe to go up to 30 seconds for a comm loss delay).

You might already know all this, but other readers may not, so forgive me if I am stating things that are obvious.

I downloaded the FC51 book and searched Modbus, and I think the comm timeout setting is parameter 8-36 and the default is 5 seconds. They call it Max Response Delay, which is in my opinion misleading since I think of "response delay" as what they describe in the previous parameter which has to do with how fast the VFD replies to a message. Anyhow, try turning that up to the max and see if it changes the behavior.