Too bad I don't have a proper answer, since I can only think of shakemaps for earthquakes occurring near the coastline - i.e. estimates are automatically computed anyway in those cases, although I doubt those estimates can take different attenuation models in account.. It'a fascinating issue..

I feel if we are having recording stations near the coast line we can record the PGA. Recently in Japan K-Net has recorded the PGA at about 273 stations for an offshore earthquake. Richard has rightly mentioned that if you search through google you will find many references.

http://www.kyoshin.bosai.go.jp/kyoshin/docs/overview_kyoshin_index_en.html

Here's the K-Net site. All their sensors appear to be on land.

I presume you are interested primarily in the vertical component. I fear the acceleration measurable at land sites will necessarily depend on the horizontal component.