In general, the following methods help but I can see that you may have just passed a few stages by now.
1. Get a strong opinion leader in the field to help you choose the right panelist and enlist their help in contacting them.
2. Make an initial contact, preferably by phone before mailing the questions. My advice is that you can still do so even after mailing the questions. The response rate will improve more significantly than if you sent a reminder by email.
3. Use of a modified Delphi (closed ended questions) instead of the traditional open ended questions for the initial round also helps in improving response rates.
4. If all fails you can always use the tried and tested method of incentives (?bribes). This can either be financial or at best material. You must, however, be very creative especially with your material resources.
Be very careful with "incentives" - use of them will need to be approved by the ethics committee that approved the study. I often find that using a group (a support group/organisation/interest group) which is central to the population that you are targeting can often help spread the word of the study and in turn assist with recruitment numbers.
I agree with you Kathryn that incentives need approval from a IRB or ethics committee and may affect the validity of one's study by introducing bias. Community involvement is also key to success of most research projects. The delphi technique is a bit different from the usual research techniques in that your "study participants" are not co-located in any community. they are in a virtual world. It involves contacting "experts"in a particular field usual through post or email to get some consensus on a particular matter. These experts may be scattered all over the world and stand to gain nothing from your study. You need them to assist you and you have to do what it takes to get their responses.
I am aware of the Delphi technique - however the use of "incentives" and bribes to gain recruits to a study would always need to be ethically approved. It is likely to have a large impact on the validity of the data being collected due to the likelihood of participant bias.