SOA at the data level?
The advantage of applying SOA at data layer, is manyfold. Primarily it offers a mechanism to streamline data access across heterogeneous sources, which was earlier done with quite cumbersome custom mechanisms. Further, it brings about a uniform interface to carry out shared tasks like common schema/data definition, shared security etc. for diverse data sources. An important effect is also in seen in form of a more regulated pattern of entry / exit to data sources, which in conventional deployments would be a mess of point to point connections. For more on shared data services and their role in SOA see the following articles:
1. Leveraging shared data services in data integration, by Krishnendu Kunti, Mohit Chawla, Vikram Sitaram, in SETLabs briefings, Vol 1, Service Oriented Architecture, 2007, pp 3-8 , accessible http://www.infosys.com/technology/toc-soa-managing-hype.asp or http://www.infosys.com/technology/setlabs-briefings-soa.pdf
2. International Conference on Next Generation Web Services Practices (NWeSP'06) pp. 49-55, An approach to automating transactions in a data services platform, Srinivas padmanabhuni et al. Accessible http://doi.ieeecomputersociety.org/10.1109/NWESP.2006.7
3. V. Niranjan, Sriram Anand, Krishnendu Kunti: Shared Data Services: An Architectural Approach. ICWS 2005: pp 683-690, Accessible http://ieeexplore.ieee.org/xpl/freeabs_all.jsp?tp=&arnumber=1530861&isnumber=32665


