You can average the 3 plates.  Sounds totally legit to me, as long as you keep records explaining what you did and why. 

Now you know the right dilution, u can prepare 3 plates with this dilution and then count the colonies and calculate average

First, I am confused about your samples.  Do you take three different samples from each site and then dilute all three?  I would recommend taking one sample, then doing a serial dilution (ie sample 1 ml of water with no dilution, the 1 ml of a 1 in 9 dilution, then 1 ml of a 1 in 99 dilution.  If you still are getting low counts, use larger plates and a greater volume of sample (the 30-300 count is for a 100mm plate).  You can also use half a ml (or 1/4) of the undiluted sample to see if that gives you countable numbers.

If you have enough capacities I would take three samples, prepare a dilution series of each sample and plate three appropriate dilutions. If one ml of the undiluted sample gives not enough colonies you can filtrate 10 or 100 ml of the sample and incubate the filter. According to the German tab water regulations you have to filtrate 100 ml. I am very sure there are similar international regulations and plenty of literature to get further information.

If you had taken three samples from the same site then there must be correlation among the samples. So it is advisable that your all the three samples must contain the similar pattern of result but with slight difference. Then dilute each sample individually by making ten fold serial dilution. After plates count the number of colonies of each individual dilution and take the average of these. It would be exact CFU/ml.