After struggling a little, I found kind of a solution. I have to admit that the solution seems a bit awkward. It works at least for the 6.0.1 release of RTC.

For the first problem I had (downloading SCM attributes), it is possible to retrieve the information as follows.

IProcessClientService processClientService =
    (IProcessClientService) this.teamRepository.getClientLibrary(IProcessClientService.class);
IProjectArea area = (IProjectArea) this.teamRepository.itemManager().fetchCompleteState(projectAreaHandle, null);
IProcessConfigurationData configData =
    processClientService.getClientProcess(area, null).getProjectConfigurationData(
        ExtendedAttributeConstants.XML_CONFIGURATION_DATA_ID, null);

(where the constant XML_CONFIGURATION_DATA_ID is set to "com.ibm.team.scm.service.extendedAttributeDefinition")

With this configData you can do something like this:

for (IProcessConfigurationElement element : configData.getElements()) {
  // ...
  String extractedValue = element.getAttribute(xmlAttributeName);
}

(xmlAttributeName being the attribute you are interested in, e.g. "extendedAttributeDefinitionKey" for the name)

Now my second problem seems to be significantly more difficult to address. The only solution I could think of is this.

IProjectArea area = (IProjectArea) this.teamRepository.itemManager().fetchCompleteState(projectAreaHandle, null);
IProcessItemService processItemService =
    (IProcessItemService) this.teamRepository.getClientLibrary(IProcessItemService.class);
IProjectArea workingCopy = (IProjectArea) processItemService.getMutableCopy(area);
@SuppressWarnings("rawtypes")
Map procData = area.getProcessData();
IContent procContent = (IContent) procData.get(ProcessContentKeys.PROCESS_SPECIFICATION_KEY);
ByteArrayOutputStream outStream = new ByteArrayOutputStream();
this.teamRepository.contentManager().retrieveContent(procContent, outStream, null);
String xmlContent = outStream.toString(IContent.ENCODING_UTF_8);
Document xmlSpec = convertStringToDocument(xmlContent);
String updatedXml = buildUpdatedXML(xmlSpec, definitions);
IContent newContent =
    this.teamRepository.contentManager().storeContent(IContent.CONTENT_TYPE_XML, updatedXml, null);
workingCopy.getProcessData().put(ProcessContentKeys.PROCESS_SPECIFICATION_KEY, newContent);
processItemService.save(workingCopy, null);

where buildUpdatedXML replaces a node in the XML document:

private String buildUpdatedXML(final Document xmlSpec, final List definitions) {
    Element element = findExtendedAttributesDOMElement(xmlSpec);
    element.getParentNode().replaceChild(xmlSpec.importNode(generateDOMNode(definitions, element), true), element);
    return convertDocumentToString(xmlSpec);
}

The method findExtendedAttributesElement() does some traversal on the XML DOM. It finds or creates some parent node in the XML to which the new content (attribute definition) could be added. The method generateDOMNode(...) generates a new DOM node based on the attribute definition I would like to put on the server. The classes for attribute definitions (ExtendedAttributeDefinition) is my own POJO for holding some attributes of the attribute definition. The name is inspired by some class deep in the SDK sources (from which I took the general idea for part 1 of my solution). 

To me, this solution does not seem to be a very clean and stable one. Especially as I am using plain XML replacement that was considered as bad practice in several other discussions here.

So I would be grateful if anyone having a better solution to this could share it with me.